Equipment state intelligent early warning method based on danger perception

Abstract

The invention provides an equipment state intelligent early warning method based on danger perception, and relates to the technical field of equipment state intelligent early warning, and the method comprises the steps: obtaining equipment operation data which comprises multi-source heterogeneous data; performing feature extraction on the multi-source heterogeneous data to obtain unified feature representation; according to the unified feature representation, constructing a correlation model between the equipment operation state and the feature parameters; and obtaining to-be-analyzed equipment operation data. According to the method, an evaluation index system is automatically generated, an evidence reasoning network is adopted for comprehensive evaluation, online incremental learning and a self-adaptive threshold adjustment mechanism are introduced, real-time and accurate evaluation of the health state of the equipment is achieved, the method effectively fuses multi-source heterogeneous data, expert knowledge and a deep learning technology, and the method is suitable for large-scale popularization and application. The comprehensiveness, accuracy and real-time performance of equipment health state evaluation are improved, and powerful support is provided for equipment predictive maintenance.

Description

Technical Field

[0001] The present invention relates to the field of intelligent early warning of equipment status, and in particular to an intelligent early warning method of equipment status based on danger perception. Background Art

[0002] In the process of equipment health status assessment and early warning, there is a technical contradiction between the multi-dimensional heterogeneity of data and the complexity of domain knowledge. On the one hand, the data generated during the operation of the equipment has the characteristics of multi-source heterogeneity, high dimension, and nonlinearity. The correlation of different types of data in time and space is intricate, and the feature extraction and fusion of data face huge challenges. On the other hand, the assessment and early warning of equipment health status needs to integrate the experience and knowledge of domain experts, build a reasonable evaluation index system, and organically combine qualitative domain knowledge with quantitative data analysis results to form an interpretable and trustworthy early warning result. In addition, in practical applications, the working conditions of equipment are changeable and the external environmental factors are complex and diverse. How to adjust the data association analysis model and early warning model in real time in a dynamically changing environment to ensure the accuracy and real-time nature of the early warning results is also a technical problem that needs to be solved urgently. At the same time, the health status assessment and early warning of equipment need to consider the entire life cycle of the equipment. How to mine valuable information from massive historical data and build a comprehensive and accurate health status portrait is also a challenging problem. Summary of the invention

[0003] The present invention provides an intelligent early warning method for equipment status based on danger perception, which mainly includes:

[0004] Acquire equipment operation data, wherein the equipment operation data includes multi-source heterogeneous data; perform feature extraction on the multi-source heterogeneous data to obtain a unified feature representation; construct a correlation model between equipment operation status and feature parameters based on the unified feature representation; acquire equipment operation data to be analyzed, analyze the equipment operation data to be analyzed based on the correlation model, and obtain equipment operation status analysis results; evaluate the current health status of the equipment based on the equipment operation status analysis results and in combination with a preset equipment health status evaluation indicator system to obtain equipment health status evaluation results.

[0005] Furthermore, the feature extraction of the multi-source heterogeneous data to obtain a unified feature representation includes: acquiring the multi-source heterogeneous data, and processing the multi-source heterogeneous data using data cleaning and preprocessing methods; for the processed multi-source heterogeneous data, using a principal component analysis method to reduce the dimension and extract key feature parameters; inputting the extracted key feature parameters into an autoencoder network, and obtaining a unified feature representation of the multi-source heterogeneous data through unsupervised learning.

[0006] Furthermore, the association model between the equipment operation status and the characteristic parameters is constructed according to the unified feature representation, including: obtaining historical equipment operation data, and extracting the characteristic parameters of the historical equipment operation data according to the unified feature representation; obtaining the equipment operation status label corresponding to the historical equipment operation data; using the characteristic parameters as input and the equipment operation status label as output, training a support vector machine model to obtain the association model between the equipment operation status and the characteristic parameters.

[0007] Furthermore, according to the equipment operation status analysis result, combined with a preset equipment health status evaluation index system, the current health status of the equipment is evaluated to obtain an equipment health status evaluation result, including: obtaining a preset equipment health status evaluation index system, the index system including multiple evaluation dimensions and corresponding index items; in combination with the equipment operation status analysis result, extracting index items related to the current equipment operation condition from the index system as evaluation indicators; for each evaluation indicator, judging the health status of the indicator according to its threshold range and actual value; comprehensively considering the health status of each evaluation indicator, and adopting a weighted average method to obtain the current health status evaluation result of the equipment.

[0008] Furthermore, the method also includes: acquiring real-time data during the operation of the equipment, and using an online incremental learning method to dynamically update the association model to adapt to changes in the equipment operating conditions; predicting the equipment operating status based on the updated association model, and when the prediction result exceeds a preset threshold, triggering an early warning and generating an alarm message.

[0009] Furthermore, before constructing the association model between the equipment operating status and the characteristic parameters, it also includes: using a frequent pattern mining algorithm for the unified feature representation to discover association rules and time series patterns in multi-source heterogeneous data; using the association rules and time series patterns obtained by mining as prior knowledge to guide the construction of the association model, thereby improving the expression ability and generalization performance of the model.

[0010] Furthermore, after obtaining the current health status evaluation result of the equipment by comprehensively considering the health status of each evaluation indicator using the weighted average method, the method also includes: obtaining equipment operation data, continuously collecting real-time data from various sensors, and performing data cleaning and preprocessing to remove outliers and noise; comparing the equipment health status evaluation result with a preset health level threshold to determine the current health level of the equipment; and automatically generating equipment maintenance priority plans and maintenance strategies based on the health level and equipment importance to guide the development of equipment operation and maintenance work.

[0011] Furthermore, a health status assessment index system is obtained, including: obtaining multiple indicators as evidence nodes; for the index system, using a hierarchical analysis method to determine the weight of each evidence node, and using the weight as the credibility of the evidence node; calculating the Pearson correlation coefficient between different evidence nodes, and using the correlation coefficient as the correlation strength between the evidence nodes; constructing an evidence reasoning network based on the credibility of the evidence nodes and the correlation strength; in the evidence reasoning network, using a Bayesian network reasoning algorithm to comprehensively analyze the support of each evidence node for different health states.

[0012] Furthermore, the acquisition of equipment operation data includes: preprocessing the equipment operation data, removing outliers and noise therein, and filling in missing values; setting a sliding time window according to the type and working characteristics of the equipment, and slicing and aggregating the preprocessed equipment operation data within the sliding time window; extracting time domain features and frequency domain features of the aggregated data, wherein the time domain features include mean, variance and peak-to-peak value, and the frequency domain features include spectrum entropy; using the Pearson correlation coefficient and the maximum information coefficient, performing importance analysis on the extracted features, and screening out the feature subset with the greatest correlation with the equipment failure state; using a support vector machine to establish a classification model between the feature subset and the equipment failure mode; acquiring The weight coefficient corresponding to each feature in the classification model is used to determine the feature with a weight coefficient greater than a preset threshold as the key factor causing equipment failure; the equipment operation status is monitored online, and the root mean square value and kurtosis of the currently collected vibration signal are calculated; if the root mean square value and kurtosis exceed the preset range, it is determined that the equipment operating condition has changed significantly, and the length of the sliding time window is dynamically shortened; the newly collected operation data of the equipment is added to the training set, and the classification model is incrementally updated using the online sequential extreme learning machine algorithm; based on the updated classification model, the remaining service life of the equipment is predicted; if the remaining service life is lower than the preset threshold, the equipment importance, maintenance cost and failure consequence factors are combined to generate an equipment maintenance plan through a multi-objective optimization algorithm.

[0013] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0014] The present invention discloses an intelligent early warning method for equipment status based on hazard perception. In view of the multi-source heterogeneous and high-dimensional nonlinear characteristics of equipment operation data, a unified representation model is constructed by multi-perspective data fusion technology to achieve standardized representation of heterogeneous data. Based on this, the temporal and spatial relationships between parameters are analyzed by the spatiotemporal association mining algorithm to discover implicit patterns. Based on this, high-level features are extracted by deep learning methods, and a domain knowledge base is constructed in combination with knowledge graph technology. The present invention automatically generates an evaluation index system, uses an evidence reasoning network for comprehensive evaluation, and introduces online incremental learning and adaptive threshold adjustment mechanisms to achieve real-time and accurate evaluation of the health status of the equipment. This method effectively integrates multi-source heterogeneous data, expert knowledge and deep learning technology, improves the comprehensiveness, accuracy and real-time performance of equipment health status evaluation, and provides strong support for predictive maintenance of equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The present invention is a flow chart of an intelligent early warning method for equipment status based on danger perception. DETAILED DESCRIPTION

[0016] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] like Figure 1 In this embodiment, a device status intelligent early warning method based on danger perception may specifically include:

[0018] Step S101, in view of the multi-source heterogeneous characteristics and high-dimensional nonlinear characteristics of equipment operation data, multi-view data fusion technology is used to build a unified representation model for heterogeneous data, map data from different sources and types to the same feature space, and realize standardized representation and fusion processing of data.

[0019] According to the multi-source heterogeneous characteristics of equipment operation data, data preprocessing technology is used to clean and convert data from different sources and types, eliminate noise and redundancy in the data, and improve data quality. In view of the high-dimensional nonlinear characteristics of the data, the principal component analysis (PCA) algorithm is used to extract the most representative and discriminative low-dimensional features from the high-dimensional data, reduce the data dimension, and reduce the computational complexity. A unified representation model for heterogeneous data is constructed, and multi-view learning technology is used to map data from different sources and types to the same feature space to achieve standardized representation of the data. Specifically, an autoencoder can be used to extract and compress the data from each perspective, and then the feature vectors of different perspectives are spliced ​​together to form a unified data representation. On the basis of the unified representation of heterogeneous data, a weighted average data fusion algorithm is used to fuse the data representations of multiple perspectives to obtain a comprehensive data representation, thereby improving the integrity and consistency of the data. Among them, the weight of each perspective can be determined according to its influence on the equipment operation status, and the perspective with greater influence has a higher weight. In view of the time series characteristics of equipment operation data, time series analysis methods such as sliding windows or recurrent neural networks (RNN) are used to convert data in the time dimension into feature vectors, which are input into the machine learning model together with other features. Based on the fused data representation and time series characteristics, the support vector machine (SVM) algorithm is used to build a classification model for the equipment operation status, and intelligent early warning and fault diagnosis of equipment status based on danger perception are realized. In view of the parameter tuning problem in the model training process, the grid search method is used to automatically search for the optimal model hyperparameter combination within the preset parameter range to improve the generalization performance and robustness of the model. Through cross-validation, the performance of the model under different parameter combinations is evaluated, and the best performing parameters are selected as the final model parameters. In the process of data collection, transmission and storage, security measures such as data encryption and access control are used to protect the confidentiality and integrity of data and prevent data leakage and tampering. At the same time, relevant data privacy protection specifications and standards, such as GDPR, are followed to ensure data compliance and user privacy protection. The trained model is deployed to the production environment to predict and analyze the real-time collected equipment operation data, and the results are presented to users through a visual interface to assist users in equipment operation and maintenance and decision-making, thereby improving equipment operation efficiency and reliability.

[0020] For example, multi-source heterogeneous equipment operation data, such as from sensors (temperature, pressure, vibration), controllers (switch status, operating parameters), log files (event records, error messages), etc., have different data formats, sampling frequencies, and dimensions. In order to effectively use these data, data preprocessing is required. For example, the data collected by the temperature sensor is in degrees Celsius, and the data collected by the pressure sensor is in Pascal. They need to be converted into a unified unit or standardized to eliminate the influence of the dimension. In addition, there may be noise in the data, such as fluctuations in sensor readings, and a filtering algorithm (such as Kalman filtering) is needed to remove the noise. At the same time, some data may be redundant, such as multiple sensors measuring the same physical quantity, and feature selection or dimensionality reduction processing is required to reduce data redundancy. This can improve data quality and lay the foundation for subsequent analysis. Equipment operation data usually has high-dimensional nonlinear characteristics. For example, a device may have dozens of sensors, and the data collected by each sensor constitutes a dimension. There may be complex nonlinear relationships between these data. In order to reduce computational complexity and avoid dimensionality disasters, the principal component analysis (PCA) algorithm can be used. PCA can project high-dimensional data into a low-dimensional space while retaining the main variance of the data. For example, reducing the dimension of 50-dimensional sensor data to 5 dimensions while retaining 95% of the data variance not only reduces the data dimension but also retains the key information of the data. In order to fuse data from different sources and different types together, it is necessary to build a unified representation model for heterogeneous data. Multi-view learning technology can be used to treat data from different sources as different views. For example, sensor data is one view, controller data is another view, and log data is a third view. The data of each view can be feature extracted and compressed using an autoencoder. Assume that the sensor data is compressed into a 10-dimensional feature vector, the controller data is compressed into a 5-dimensional feature vector, and the log data is compressed into a 2-dimensional feature vector after the autoencoder. These three feature vectors are spliced ​​together to form a 17-dimensional unified data representation. This can map different types of data to the same feature space, which is convenient for subsequent fusion and analysis. After obtaining a unified data representation, a weighted average data fusion algorithm can be used. For example, based on expert experience or domain knowledge, it is determined that the influence of sensor data, controller data, and log data on the equipment operation status is 0.6, 0.3, and 0.1 respectively. Then, the feature vectors of these three views are multiplied by the corresponding weights and added together to obtain a comprehensive 17-dimensional data representation. This weighted fusion method can reflect the importance of different data sources and improve data integrity and consistency. Equipment operation data usually has time series characteristics. For example, the data collected by the sensor changes over time to form a time series. You can use a sliding window to convert time series data into a feature vector.For example, taking the data of 10 time points as a window, the mean, variance, maximum value, minimum value and other statistical features of the data in the window are calculated to form a feature vector. You can also use a recurrent neural network (RNN), such as LSTM, to model time series data and extract time series features. These time series features are input into the machine learning model together with other features. Using the fused data representation and time series features, the support vector machine (SVM) algorithm can be used to build a classification model for the equipment operation status. For example, based on historical data, the equipment operation status is divided into three states: normal, abnormal, and fault. An SVM classifier is trained to classify the new equipment operation data to achieve fault diagnosis. During the model training process, the model parameters need to be tuned. The grid search method is used to automatically search for the optimal model hyperparameter combination within the preset parameter range, such as the kernel function type and penalty coefficient of the SVM. Through cross-validation, such as 5-fold cross-validation, the performance of the model under different parameter combinations is evaluated, and the best performing parameters are selected as the final model parameters to improve the generalization performance and robustness of the model. In order to ensure data security, security measures such as data encryption and access control need to be taken. For example, data is encrypted for storage and transmission, and access rights to data are restricted to ensure that only authorized users can access data to prevent data leakage and tampering. At the same time, relevant data privacy protection specifications and standards need to be followed to ensure data compliance and user privacy protection. Finally, the trained model is deployed to the production environment to predict and analyze the real-time collected equipment operation data, and the results are presented to the user through a visual interface. For example, the predicted equipment operation status, fault type and other information are displayed on the monitoring platform to assist users in equipment operation and maintenance and decision-making, and improve the equipment's operating efficiency and reliability.

[0021] Step S102: Based on the uniformly represented data, a spatiotemporal correlation mining algorithm is used to analyze the temporal relationship and spatial dependency between the equipment operating parameters to discover the implicit correlation patterns and evolution laws in the data. Specifically, a sliding time window and spatial correlation analysis are used, combined with a graph theory algorithm to construct a parameter correlation network, thereby identifying key spatiotemporal patterns.

[0022] Based on the unified representation of equipment operation data, a multidimensional spatiotemporal data model is constructed to map the operating parameters of different equipment into a unified spatiotemporal coordinate system. The sliding time window technology is used to dynamically divide time slices, extract the statistical characteristics of the equipment operation parameters in each time slice, such as mean, variance, etc., to form a feature sequence in the time dimension. Based on spatial correlation analysis methods such as the Pearson correlation coefficient, the spatial correlation coefficient matrix between the operating parameters of different equipment is calculated to characterize the correlation strength of the equipment in the spatial dimension. The feature sequence in the time dimension and the correlation coefficient matrix in the spatial dimension are integrated to construct a spatiotemporal correlation feature matrix to comprehensively characterize the spatiotemporal evolution law of the equipment operation status. Frequent pattern mining algorithms such as the Apriori algorithm are applied to find frequently co-occurring parameter combination patterns in the spatiotemporal correlation feature matrix to reveal the key spatiotemporal correlation rules of equipment operation. Combined with graph theory algorithms such as the minimum spanning tree algorithm, the equipment operation parameters are used as nodes and the spatiotemporal correlation strength is used as the edge weight to construct an equipment operation parameter association network to intuitively display the spatiotemporal dependency between parameters. Through community discovery algorithms such as the Louvain algorithm, closely related parameter clusters are identified in the parameter association network, revealing the key spatiotemporal patterns of equipment operation and providing decision support for equipment operation and maintenance optimization.

[0023] Exemplarily, the construction of a multidimensional spatiotemporal data model is to map the operating parameters of different devices into a unified spatiotemporal coordinate system, so as to analyze the spatiotemporal correlation between devices. Assume that there are multiple CNC machine tools in a factory workshop, and each machine tool is equipped with multiple sensors for monitoring parameters such as temperature, pressure, and vibration. Each machine tool can be regarded as a point in space, and each time point can be regarded as a scale in the time dimension, so as to construct a three-dimensional spatiotemporal coordinate system. In order to extract the feature sequence in the time dimension, a sliding time window technology is used. For example, a time window of 10 minutes is used, and it slides every 5 minutes. In each time window, the statistical features of each parameter of each machine tool are calculated, such as mean, variance, etc. Assume that the temperature mean of machine tool A in the first time window (0-10 minutes) is 25°C and the variance is 1°C; the temperature mean in the second time window (5-15 minutes) is 26°C and the variance is 0.5°C. In this way, the feature sequence of the temperature parameters of machine tool A in the time dimension can be formed. Spatial correlation analysis is used to calculate the spatial correlation coefficient matrix between the operating parameters of different equipment and characterize the correlation strength of the equipment in the spatial dimension. For example, the Pearson correlation coefficient between the temperature of machine tool A and the pressure of machine tool B is calculated. Assuming the correlation coefficient is 0.8, it indicates that there is a strong positive correlation between the temperature of machine tool A and the pressure of machine tool B, that is, when the temperature of machine tool A increases, the pressure of machine tool B also increases. By calculating the correlation coefficients between all equipment operating parameters, a spatial correlation coefficient matrix can be obtained. By fusing the feature sequence in the time dimension and the correlation coefficient matrix in the spatial dimension, a spatiotemporal correlation feature matrix can be constructed. For example, the mean and variance of the temperature of machine tool A in each time window, as well as the correlation coefficient of the temperature of machine tool A with other equipment parameters are integrated together to form the spatiotemporal correlation feature vector of machine tool A. Combining the spatiotemporal correlation feature vectors of all equipment together constitutes a spatiotemporal correlation feature matrix. Frequent pattern mining is used to discover frequently co-occurring parameter combination patterns in the spatiotemporal correlation feature matrix. For example, by applying the Apriori algorithm, it can be found that the combination pattern of "the temperature of machine tool A increases, the pressure of machine tool B increases, and the vibration of machine tool C increases" appears frequently. This reveals that there are spatiotemporal association rules between these three parameters. Constructing an equipment operation parameter association network can intuitively show the spatiotemporal dependency between parameters. A parameter association network is constructed with equipment operation parameters as nodes and spatiotemporal association strength (for example, the absolute value of the correlation coefficient) as edge weight. For example, if the correlation coefficient between the temperature of machine tool A and the pressure of machine tool B is 0.8, the temperature node of machine tool A and the pressure node of machine tool B are connected in the network with an edge weight of 0.8. The community discovery algorithm is used to identify closely related parameter clusters in the parameter association network. For example, by applying the Louvain algorithm, it can be found that there is a parameter cluster in the network that includes the temperature of machine tool A, the pressure of machine tool B, and the vibration of machine tool C.This indicates that there is a strong spatiotemporal dependency between the three parameters, and they may be affected by a common underlying factor. Identifying these key spatiotemporal patterns can help engineers better understand the operating rules of the equipment and optimize equipment operation and maintenance. For example, if a certain parameter cluster is found to be frequently abnormal, it can be monitored and maintained in a focused manner to improve the reliability and operating efficiency of the equipment.

[0024] Step S103, using the discovered spatiotemporal correlation patterns as input for feature engineering to guide the construction of a multi-level feature extraction network. The deep learning method is used to automatically learn high-level feature representations in the data, including using convolutional neural networks to capture local features and long short-term memory networks to model temporal dependencies, effectively reducing data dimensions and improving the efficiency and accuracy of feature extraction.

[0025] According to the spatiotemporal correlation pattern, a feature engineering framework is constructed, and the spatiotemporal correlation pattern is used as the input of feature engineering. By analyzing the spatiotemporal correlation pattern, key features such as the trend, periodicity, and spatial distribution of the time series are extracted as the input of the subsequent feature extraction network. A multi-level network structure combining a convolutional neural network (CNN) and a long short-term memory network (LSTM) is used to design a feature extraction network. The convolution operation of CNN is used to capture the local features and spatial correlation of the data, and features are extracted and abstracted layer by layer through multi-layer convolution and pooling operations. At the same time, an LSTM module is embedded in the network, and the temporal dependency of the data is modeled through the gating mechanism and memory unit to capture the evolution law of the data in the time dimension. In the feature extraction process, dimensionality reduction techniques such as principal component analysis (PCA) are used to reduce the dimensionality of high-dimensional data and reduce redundant and noise features. At the same time, feature selection methods such as mutual information and correlation coefficient are used to select the most discriminative and relevant feature subsets to improve the efficiency and accuracy of feature extraction. The feature extraction network is trained and optimized end-to-end using the back propagation algorithm. By defining a suitable loss function, such as mean square error (MSE) or cross entropy loss, the difference between the network output and the true label is calculated, and the network parameters are updated using the gradient descent method, so that it can adaptively learn and extract high-level feature representations from the data. At the output layer of the feature extraction network, the high-level feature representation of the data is obtained. The extracted high-level features are input into subsequent task models, such as classifiers or regression models such as support vector machines (SVM) and random forests, for tasks such as classification, prediction, or anomaly detection. The high-level features automatically learned and extracted by the feature extraction network can effectively improve the performance and accuracy of the task. In practical applications, the structure and parameters of the feature extraction network are adjusted and optimized according to the requirements of specific tasks. Through methods such as cross-validation, the optimal network structure and hyperparameter settings are selected to achieve the best feature extraction effect. At the same time, data enhancement techniques such as rotation, translation, and scaling are used to expand the training data set to improve the generalization ability and robustness of the feature extraction network.

[0026] For example, the spatiotemporal correlation pattern can be used as the input of feature engineering to provide strong data support for subsequent machine learning tasks. The core goal of the feature engineering framework is to extract valuable features from the raw data in order to better understand and model the data. First, the spatiotemporal correlation pattern needs to be analyzed. For example, suppose that the temperature, pressure and vibration data collected by multiple sensors on a factory assembly line are obtained, and the spatiotemporal correlation mining algorithm is used to find the frequently occurring pattern of "the temperature of device A increases, the pressure of device B increases, and the vibration of device C increases". By analyzing this pattern, the following key features can be extracted: the time series trend of the parameters of devices A, B, and C (for example, the temperature continues to rise, the pressure fluctuation increases, and the vibration amplitude increases), the periodicity (for example, whether there is an obvious daily cycle or weekly cycle in the parameter change), and the spatial position distribution of devices A, B, and C (for example, whether the three devices are adjacent to each other on the assembly line). These features can more carefully characterize the operating status of the equipment and their mutual influence. Next, the extracted key features are used as input to construct a feature extraction network. The network usually adopts a multi-level structure combining CNN and LSTM. CNN is good at capturing the local features and spatial correlation of data. For example, CNN can learn the impact of the temperature increase of device A on the pressure of device B in space. Through multi-layer convolution and pooling operations, the network can extract and abstract features layer by layer, such as extracting more representative features such as "temperature rise rate", "pressure fluctuation frequency", and "vibration peak" from the original temperature, pressure, and vibration values. LSTM is used to model the temporal dependencies of data. For example, LSTM can learn that the pressure of device B will increase only after the temperature of device A increases for a period of time, thereby capturing the temporal evolution law between parameters. In the feature extraction process, in order to reduce data redundancy and noise, dimensionality reduction techniques such as PCA can be used. For example, assuming that 10 features are extracted, but PCA analysis shows that only 3 principal components can explain 95% of the data variance, then the feature dimension can be reduced from 10 to 3, simplifying the training of subsequent models. At the same time, feature selection methods such as mutual information or correlation coefficient can be used to further screen out the features most relevant to the target task. For example, if it is found that the "temperature rise rate" is most strongly associated with device failure, it can be used as the main feature input into the subsequent fault prediction model. The training of the feature extraction network uses the back propagation algorithm. For example, if the target task is to predict equipment failure, the mean square error can be defined as the loss function to measure the difference between the network's prediction and the actual fault label. The network parameters are continuously updated through the gradient descent method, so that the network can learn the optimal feature representation, thereby improving the prediction accuracy. After training, the output layer of the feature extraction network will provide a high-level feature representation of the data. These high-level features can be input into various task models. For example, SVM can be used to classify equipment failures, or random forests can be used to predict the remaining life of equipment.Compared with directly using raw data, using extracted high-level features can significantly improve the performance and accuracy of the task model. In practical applications, the feature extraction network needs to be adjusted and optimized according to specific task requirements. For example, for image data, a deeper CNN structure can be used; for time series data, the number of hidden units of LSTM can be adjusted. The optimal network structure and hyperparameter settings can be determined by methods such as cross-validation. In addition, data enhancement techniques, such as random cropping or adding noise to time series, can effectively improve the generalization ability and robustness of the network, so that it can maintain good performance when facing new data. In summary, the feature engineering framework based on spatiotemporal correlation patterns can effectively extract high-level features from data by combining deep learning technologies such as CNN and LSTM, as well as PCA, feature selection and other methods, providing strong support for various downstream tasks.

[0027] Step S104, combining the extracted high-level features, using knowledge graph technology to build a domain knowledge base for equipment health status assessment, formalizing expert experience into a knowledge graph, and realizing structured representation and reasoning of domain knowledge. The knowledge graph contains entities such as equipment components, fault types, performance indicators, and their relationships.

[0028] The performance indicator data of the equipment during operation are obtained and preprocessed, including data cleaning, outlier processing and data normalization, to obtain a standardized indicator data set. According to the pre-built equipment performance knowledge graph, the indicator name in the standardized indicator data set is matched with the performance indicator entity name in the knowledge graph to obtain the entity and its attribute information related to the current equipment operation status. According to the obtained performance indicator entity attribute information, the real-time value of each performance indicator is calculated. Through knowledge reasoning, the real-time value of the performance indicator is compared with the preset normal range threshold to determine whether it exceeds the normal range. If it exceeds the normal range, the fault type entity related to the performance indicator is queried in the knowledge graph. Knowledge reasoning can use rule-based reasoning methods, such as defining a series of IF-THEN rules, triggering corresponding rules according to the comparison results of the real-time value of the performance indicator and the threshold, and inferring the possible fault type. Combined with the associated information of the fault type entity in the knowledge graph, the equipment component entity that may cause the fault is further obtained to form a preliminary evaluation result of the current health status of the equipment. The entity attribute information extracted from the knowledge graph is used as high-level features, and the decision tree algorithm is used to analyze the high-level features. The importance of each feature is evaluated through indicators such as information gain or Gini index, and key features that have an important impact on the judgment of the equipment health status are selected. Considering the preliminary evaluation results and key feature analysis results, the current health status assessment report of the equipment is output. The report is generated in text format, including basic equipment information, current health status summary, possible fault types, equipment components involved in the fault, and analysis of the causes of the fault, etc., to provide decision support for subsequent maintenance and repair.

[0029] For example, the health management of equipment is like a physical examination of the human body. In order to fully understand the health status of the equipment, it is necessary to collect various performance indicator data during the operation of the equipment, such as temperature, pressure, vibration, current, etc. These data are like various physiological indicators of the human body, such as body temperature, blood pressure, heart rate, etc. After obtaining the data, it is necessary to preprocess it first, just like some preparation work needs to be done before a physical examination. Data cleaning is to remove noise and errors in the data, such as removing abnormal spikes or missing values. Outlier processing is to correct or eliminate data that exceeds the normal range, such as the temperature at a certain point in time suddenly soaring to an impossible value. Data normalization is to convert data of different dimensions to the same scale range, such as converting temperature values ​​from Celsius to Fahrenheit. Through these preprocessing steps, a standardized indicator data set is obtained, laying the foundation for subsequent analysis. Suppose an industrial robot is being monitored, and its joint temperature, motor current, and movement speed are collected. After data cleaning, the abnormal spikes caused by sensor failure are removed; through outlier processing, the abnormal temperature fluctuations caused by external environmental interference are corrected; after data normalization, the indicator data of different dimensions are converted to the range of 0 to 1. Next, it is necessary to use the equipment performance knowledge graph to understand these data. The knowledge graph is like a doctor's professional knowledge base, which contains various performance indicators of the equipment, fault types and the relationship between them. Match the indicator name in the standardized indicator data set with the performance indicator entity name in the knowledge graph, such as matching "joint temperature" with the "robot joint temperature" entity in the knowledge graph. In this way, the entity and its attribute information related to the current equipment operation status can be obtained, such as the normal range and alarm threshold of the "robot joint temperature" entity. For example, the normal range of the "robot joint temperature" entity in the knowledge graph is 20℃ to 40℃, and the alarm threshold is 45℃. By querying the knowledge graph, you can also obtain the fault type related to "robot joint temperature", such as "joint overheating". According to the obtained performance indicator entity attribute information, the real-time value of each performance indicator can be calculated. For example, based on the data collected by the sensor, the real-time temperature value of each joint of the robot can be calculated. Then, through knowledge reasoning, the real-time value of the performance indicator is compared with the preset normal range threshold to determine whether it exceeds the normal range. For example, the real-time temperature of joint No. 1 of the robot is 46°C, which exceeds the alarm threshold of 45°C. Through knowledge reasoning, it can be determined that the joint is at risk of overheating. If a performance indicator exceeds the normal range, the fault type entity related to the performance indicator is queried in the knowledge graph. For example, according to the knowledge graph, "robot joint temperature" exceeding the alarm threshold may cause a "joint overheating" fault. Knowledge reasoning can use rule-based reasoning methods. For example, define a rule: If the robot joint temperature exceeds 45°C, it is inferred that a "joint overheating" fault may occur.Combined with the associated information of the fault type entity in the knowledge graph, the equipment component entity that may cause the fault can be further obtained. For example, the "joint overheating" fault may be related to components such as "temperature sensor" and "cooling fan". In this way, a preliminary evaluation result of the current health status of the equipment is formed. The entity attribute information extracted from the knowledge graph is used as a high-level feature, such as "robot joint temperature", "motor current", "movement speed", etc. The decision tree algorithm is used to analyze the high-level features, and the importance of each feature is evaluated by indicators such as information gain or Gini index. For example, it is found through analysis that "robot joint temperature" has the greatest impact on the judgment of the equipment health status. Taking into account the preliminary evaluation results and the key feature analysis results, the current health status evaluation report of the equipment is finally output. The report content includes basic information of the equipment, the overall evaluation of the current health status, possible fault types, equipment components involved in the fault, and analysis of the causes of the fault. For example, the final health status assessment report is as follows: Equipment name: Industrial robot XYZ-1; Current health status: Warning; Possible fault type: Overheating of joint No. 1; Equipment components involved in the fault: Joint No. 1 temperature sensor, Joint No. 1 cooling fan; Cause analysis: The temperature of joint No. 1 continues to rise and exceeds the alarm threshold, which may be caused by a cooling fan failure or excessive ambient temperature.

[0030] Step S10.5, based on the constructed domain knowledge graph, combined with the equipment operating conditions and performance requirements, automatically generate the equipment health status evaluation index system. Specifically, through graph traversal and relationship reasoning, key evaluation indicators are identified, and the hierarchical structure in the graph is used to form a multi-dimensional evaluation system, thereby forming a quantitative evaluation method that comprehensively considers the multi-dimensional characteristics of the equipment.

[0031] According to the domain knowledge, the equipment-related knowledge graph is constructed. The knowledge graph contains multi-dimensional information such as equipment components, operating conditions, performance parameters, etc., forming a complete equipment knowledge representation. Through the ontology hierarchy in the knowledge graph, a classification system of equipment health evaluation indicators is formed, and the evaluation indicators are divided into different levels such as component level, subsystem level and system level. The breadth-first search algorithm is used to traverse the knowledge graph to identify the key nodes that are highly related to the equipment operating conditions and performance requirements, and use them as key evaluation indicators. For each key evaluation indicator, the attribute nodes directly connected to it are extracted in the knowledge graph as the influencing factors of the indicator. The real-time operation data of the equipment is obtained, and the data is mapped to the corresponding attribute nodes in the knowledge graph through the pre-established mapping table. For each attribute node, a health status threshold is set, and the health status of each indicator is judged by comparing the real-time data with the threshold. On the basis of the key evaluation indicators, the knowledge graph is traversed by the depth-first search algorithm, starting from the evaluation indicators, and the attribute nodes indirectly connected to it are mined, which are used as the influencing factors of the equipment health status to form a more comprehensive evaluation indicator system. According to the ontological hierarchical relationship between the evaluation indicators in the knowledge graph, the analytic hierarchy process is used to determine the weight of each indicator. First, a pairwise comparison matrix of indicators is constructed, and the value of each element in the matrix is ​​determined by expert scoring. Then the eigenvector corresponding to the maximum eigenvalue of the matrix is ​​calculated, and it is normalized as the indicator weight. A weighted average model is constructed as a comprehensive evaluation model, and the real-time data of each evaluation indicator is multiplied by the corresponding weight, and the sum is obtained to obtain a quantitative score of the current health status of the equipment. According to the preset health status score threshold, it is judged whether the equipment is currently in a healthy state. When the score is lower than the warning threshold, the system automatically issues a warning prompt, indicating that the equipment is at risk of failure.

[0032] For example, when constructing a knowledge graph related to a device, it is first necessary to define entities such as the components, operating conditions, and performance parameters of the device, and to clarify the relationships between them. For example, for an air-conditioning system, the entities in the knowledge graph may include components such as compressors, condensers, and evaporators, as well as their performance parameters such as operating temperature and pressure. Through the ontology hierarchy, the evaluation indicators can be classified into component level, subsystem level, and system level, thus forming a multi-level evaluation system. This classification helps to systematically understand and analyze the health status of the equipment. When traversing the knowledge graph using the breadth-first search algorithm, the key nodes that are closely related to the operating status of the equipment can be effectively identified. Taking the air-conditioning system as an example, if it is found that the operating temperature of the compressor is a key factor affecting the performance of the system, then this indicator will be selected as the key evaluation indicator. Next, by extracting attribute nodes directly related to the compressor temperature, such as coolant temperature and ambient temperature, it is possible to further analyze how these factors affect the operating efficiency of the compressor. When monitoring the operation of the equipment in real time, the collected data needs to be mapped to the corresponding attribute nodes in the knowledge graph. For example, if the real-time data shows that the operating temperature of the compressor is 50 degrees, and the normal operating temperature threshold set in the knowledge graph is 30 to 40 degrees, it can be judged that the compressor is at risk of overheating. This comparison can immediately discover potential equipment problems so that corresponding maintenance measures can be taken. Through the depth-first search algorithm, starting from a key evaluation indicator, other attribute nodes indirectly related to it can be explored. This helps to build a more comprehensive equipment health assessment system. For example, in addition to the factors that directly affect the temperature of the compressor, indirect factors such as power supply stability and compressor aging may also need to be considered. When determining the weights of each evaluation indicator, the hierarchical analysis method provides a systematic approach. First, a matrix for comparing indicators pairwise is constructed, and the element values ​​in the matrix are determined by expert scoring. The maximum eigenvalue of this matrix and its corresponding eigenvector are calculated, and then the eigenvector is normalized to obtain the weights of each indicator. These weights reflect the relative importance of different indicators in equipment health assessment. Finally, by constructing a weighted average model, the real-time data of each evaluation indicator is multiplied by the corresponding weight and summed, and the current health status quantitative score of the equipment can be obtained. For example, if the weights of compressor temperature, power stability, and compressor aging are 0.5, 0.3, and 0.2 respectively, and the current real-time data are 50 degrees, good, and medium respectively, the comprehensive score can indicate the overall health of the equipment. When the score is lower than the set warning threshold, the system will automatically issue a warning to prompt maintenance personnel to pay attention to possible equipment failures.

[0033] Step S106: For the generated evaluation index system, an evidence reasoning method is used to construct an evidence reasoning network. Expert knowledge and data analysis results are used as evidence nodes, and the Durst-Schafer theory is used to calculate the correlation and credibility between the evidences, and joint reasoning is performed to generate a health status evaluation result that comprehensively considers qualitative and quantitative factors.

[0034] According to expert knowledge and data analysis results, a health status assessment index system is constructed, and each index is used as an evidence node. The analytic hierarchy process is used to determine the weight of each evidence node through expert scoring as the credibility of the evidence node. The Pearson correlation coefficient is used to calculate the correlation between different evidence nodes as the strength of association between evidence nodes. According to the credibility and correlation strength of the evidence nodes, the network analysis software Gephi is used to construct an evidence reasoning network. In the evidence reasoning network, the Bayesian network reasoning algorithm is used to comprehensively analyze the support of each evidence node for different health states. Through Bayesian network reasoning, the posterior probability of various health states is calculated. The health state with the highest probability is the health state that the assessment object is most likely to be in, which is used as the result of the health status assessment. According to the health status assessment results, the current health level of the assessment object is analyzed and explained, and targeted health management suggestions are provided to guide the assessment object to improve its health status.

[0035] For example, to construct a health status evaluation index system, we first need to clarify the evaluation object. Assume that the evaluation object is an industrial robot and we want to evaluate its health status. Expert knowledge and data analysis results show that the health status of the robot is mainly affected by the following indicators: joint temperature, vibration amplitude, working current, working time and maintenance records. These indicators constitute the evidence nodes. The analytic hierarchy process is an effective method to determine the weight of each evidence node. Invite multiple robot experts to compare the above indicators in pairs and score them. For example, if the expert believes that the joint temperature is slightly more important than the vibration amplitude, the importance score of the joint temperature relative to the vibration amplitude is 3. By collecting the scores of all experts, a comparison matrix can be constructed. After calculation, the weight of each indicator is obtained, for example, the weight of the joint temperature is 0.3, the vibration amplitude is 0.2, the working current is 0.25, the working time is 0.15, and the maintenance record is 0.1. These weights represent the importance of each indicator in evaluating the health status of the robot. The Pearson correlation coefficient can be used to measure the strength of the association between different evidence nodes. For example, by analyzing historical data, it is found that there is a strong positive correlation between joint temperature and vibration amplitude, with a Pearson correlation coefficient of 0.8. This shows that an increase in joint temperature is often accompanied by an increase in vibration amplitude. Similarly, the correlation coefficients between other evidence nodes can be calculated, such as the correlation coefficient between working current and working time, the correlation coefficient between vibration amplitude and working current, and so on. Using network analysis software such as Gephi, an evidence reasoning network can be constructed. Each indicator is used as a node, the correlation between indicators is used as the weight of the edge, and the weight of the indicator is visualized as the size of the node. The size of the node reflects the importance of the indicator, and the thickness of the edge reflects the strength of the association between indicators. Such a network diagram can intuitively show the relationship between various indicators and help understand the influencing factors of the robot's health status. Based on the constructed evidence reasoning network, the Bayesian network reasoning algorithm can be used for health status assessment. Bayesian network is a probabilistic graph model that can effectively handle uncertainty information. First, the health status of the robot needs to be defined. For example, the health status can be divided into three states: "healthy", "sub-healthy" and "faulty". Then, based on expert knowledge and data analysis results, the support of each evidence node for different health states is determined. For example, high joint temperature may support the "fault" state, while regular maintenance records support the "healthy" state. Through Bayesian network reasoning, the posterior probabilities of various health states can be calculated. Assume that based on the current monitoring data, the probability of the "healthy" state is calculated to be 0.7, the probability of the "sub-healthy" state is 0.25, and the probability of the "fault" state is 0.05. Since the probability of the "healthy" state is the highest, the evaluation result is that the robot is currently in the "healthy" state. Finally, based on the health status evaluation results, analysis and interpretation are performed and health management suggestions are provided.For example, the robot is currently in a "healthy" state, but considering that the joint temperature has slightly increased, it is recommended to strengthen the heat dissipation and pay attention to the changes in vibration amplitude. If the evaluation result is a "sub-healthy" state, it is recommended to conduct further inspections to find out potential problems. If the evaluation result is a "faulty" state, it is necessary to shut down and repair immediately. Through the above steps, a robot health status assessment system based on evidence reasoning can be constructed. The system comprehensively considers multiple indicators and uses the Bayesian network reasoning algorithm to make a relatively accurate assessment of the robot's health status and provide corresponding health management recommendations. This helps to improve the reliability and service life of the robot and reduce maintenance costs.

[0036] Step S107, based on the credibility value and association strength of the evidence nodes, use network analysis software to build an evidence reasoning network, use the Bayesian network reasoning algorithm to comprehensively analyze the support degree of each evidence node for different health states, obtain the posterior probability of various health states, determine the most likely health state of the evaluation object, and obtain the health state evaluation result.

[0037] According to the credibility value and association strength of the evidence nodes, an evidence reasoning network is constructed. The constructed evidence reasoning network is converted into a corresponding Bayesian network model. For the constructed Bayesian network model, the Bayesian network reasoning algorithm is used to comprehensively analyze the support degree of each evidence node for different health states. According to the analysis results of the Bayesian network reasoning algorithm, the posterior probabilities corresponding to various health states are obtained. The posterior probabilities of each health state are sorted in descending order. By comparing the sorted posterior probabilities of the health states, the most likely health state of the evaluation object is determined. The posterior probability of the most likely health state is compared with the preset threshold. If it exceeds the preset threshold, the health state is output as the health state evaluation result; if it does not exceed the preset threshold, the remaining unanalyzed evidence nodes are selected according to the pre-defined evidence node priority. The selected unanalyzed evidence nodes are added to the existing evidence reasoning network, and the Bayesian network model is updated by recalculating the conditional probability table. The Bayesian network reasoning algorithm is re-executed using the updated Bayesian network model until the output condition is met or all evidence nodes have been analyzed. The health state with the highest final posterior probability is output as the health state evaluation result of the evaluation object.

[0038] For example, in practical applications, constructing an evidence reasoning network and converting it into a Bayesian network model is a common method to evaluate the state of a complex system, such as the health status assessment of an industrial robot. First, by collecting operating data about the robot, such as joint temperature, vibration amplitude, etc., these data are input into the evidence reasoning network as evidence nodes. The credibility of each evidence node is derived based on historical data and expert experience evaluation. For example, if historical data shows that abnormal joint temperature usually indicates that the robot is about to fail, the credibility of the joint temperature node will be set high. Next, based on the correlation between the evidence nodes, a Bayesian network model is constructed. For example, if the analysis finds that there is a high correlation between joint temperature and vibration amplitude, the two nodes will be connected by an edge in the Bayesian network. The weight of the edge reflects the strength of the association between the two nodes, and this weight can be determined by calculating the Pearson correlation coefficient of the two nodes. After the Bayesian network model is constructed, the Bayesian network inference algorithm is used to analyze the degree of support of each evidence node for different health states. For example, through network reasoning, it can be found that the combination of high joint temperature and high vibration amplitude greatly supports the hypothesis that the robot is in a "failure" state. Based on these analyses, the posterior probabilities of the robot being in a healthy, subhealthy, and faulty state can be calculated. By sorting the posterior probabilities, the most likely health state of the robot can be determined. If the posterior probability of the "fault" state is the highest and exceeds the preset threshold, then this state can be directly used as the health assessment result of the robot. If the probability of no state exceeds the threshold, then more evidence nodes may need to be considered, such as working current and maintenance records, and these new evidence nodes are added to the existing Bayesian network, and the conditional probability table is recalculated. In this way, the Bayesian network model can be gradually improved until a reliable health status assessment result is obtained. This Bayesian network-based reasoning method can not only provide quantitative health status assessment, but also adapt to changes in the robot's operating state by continuously introducing new evidence nodes, thereby realizing dynamic health management and preventive maintenance strategies.

[0039] Step S108: Based on the evaluation results, online incremental learning technology is used to obtain equipment operation data in real time, and the data association analysis model and early warning model are dynamically updated. Through the sliding time window and forgetting factor mechanism, the system continuously adapts to the changes in equipment working conditions. At the same time, an adaptive threshold adjustment mechanism is introduced, and the early warning threshold is dynamically adjusted using the exponentially weighted moving average method according to the equipment historical data and current operating status to improve the sensitivity and reliability of the early warning.

[0040] Obtain equipment operation data. Continuously collect real-time data from various sensors, including temperature, pressure, current, voltage, vibration frequency, etc., and perform data cleaning and preprocessing to remove outliers and noise to ensure data quality. Using the preprocessed data, the recursive least squares method is used to perform online incremental modeling of the data, and the correlation model between equipment operation parameters is updated in real time to identify the association relationship between different parameters and potential failure modes. Based on the updated data association model, the online sequence extreme learning machine is used to dynamically adjust the parameters of the fault warning model so that it can more accurately predict potential equipment failures. The online sequence extreme learning machine can achieve fast online learning by randomly generating hidden layer parameters and fixing them, and only updating the output weights. The sliding time window mechanism is used to retain only the equipment operation data within the recent period of time, so as to better reflect the current equipment working conditions and reduce the impact of historical data on the model. The time window length can be set according to the characteristics of the equipment and the data change cycle, generally a few hours to a few days. The forgetting factor mechanism is introduced to give new data a higher weight and reduce the weight of old data, so that the model can adapt to changes in equipment working conditions more quickly. The forgetting factor can be adjusted according to the speed of data change of the equipment, and the value is generally between 9 and 1. According to the historical data of the equipment and the current operating status, the exponentially weighted moving average method is used to dynamically adjust the fault warning threshold. The calculation formula is: current threshold = α × current moment prediction value + (1-α) × previous moment threshold, where α is the smoothing coefficient, which generally takes a value between 0.5 and 2. The setting of the threshold needs to comprehensively consider factors such as the safety margin, maintenance cost and false alarm rate of the equipment. If the current fault prediction value exceeds the dynamic warning threshold, a warning message is generated to notify relevant personnel to take measures. If the prediction value is lower than the threshold, it is considered that the equipment is operating normally, and online monitoring and model update continue.

[0041] Exemplarily, obtaining equipment operation data is the first step in fault warning. For example, in a chemical plant, it is necessary to monitor the temperature, pressure, current, vibration and other data of the reactor and the stirring motor. These data are collected in real time by various sensors installed on the reactor and transmitted to the data center through the industrial control network. In order to ensure data quality, the collected data needs to be cleaned and preprocessed. For example, outliers caused by sensor failure and interference caused by environmental noise are removed. The data can also be smoothed to reduce the impact of data fluctuations. Data cleaning and preprocessing can improve the accuracy and reliability of the model. Using the preprocessed data, the recursive least squares method can be used to perform online incremental modeling on the data. For example, a relationship model between the temperature and pressure of the reactor can be established. Since the operating state of the reactor changes over time, the model parameters need to be updated in real time to better reflect the current equipment operating conditions. Online incremental modeling can avoid repeated calculations of all historical data, thereby improving calculation efficiency. At the same time, it can also continuously adjust the model according to new data to improve the adaptability of the model. Based on the updated data association model, the parameters of the fault warning model can be dynamically adjusted using an online sequence extreme learning machine. For example, based on the relationship model between temperature and pressure, the temperature change trend of the reactor in the future can be predicted. The online sequence extreme learning machine is an efficient online learning algorithm. It can achieve fast online learning by randomly generating hidden layer parameters and fixing them, and only updating the output weights, so that it can adapt to changes in equipment conditions more quickly and detect potential faults in time. In order to better reflect the current equipment conditions, a sliding time window mechanism is needed. For example, only the reactor operation data of the last 24 hours is retained. This can reduce the impact of historical data on the model and make the model more focused on the current equipment status. The length of the time window needs to be set according to the characteristics of the equipment and the data change cycle. For example, for equipment that changes faster, the time window can be set shorter; while for equipment that changes slower, the time window can be set longer. The introduction of the forgetting factor mechanism can give new data a higher weight and reduce the weight of old data. For example, the forgetting factor can be set to 0.9, which means that the latest data has the greatest impact on the model, while the impact of older data on the model gradually decreases. The forgetting factor mechanism enables the model to adapt to changes in equipment conditions more quickly, thereby improving the prediction accuracy of the model. Based on the historical data of the equipment and the current operating status, the fault warning threshold can be dynamically adjusted using the exponentially weighted moving average method. For example, the temperature warning threshold can be dynamically adjusted based on the deviation between the predicted value and the actual value of the reactor temperature over a period of time. If the current temperature prediction value exceeds the dynamic warning threshold, a warning message is generated to notify relevant personnel to take measures. For example, if the reactor temperature is predicted to exceed the safety upper limit, the system will automatically issue an alarm to remind the operator to lower the reactor temperature.Dynamically adjusting the threshold can effectively avoid false alarms and missed alarms caused by changes in equipment operating conditions. For example, suppose the temperature of the reactor continues to rise and exceeds the set threshold. The system will issue an early warning message to remind the operator to check whether the reactor's cooling system is working properly. If the cooling system fails, the operator can take timely measures, such as starting the backup cooling system or reducing the operating load of the reactor to prevent the reactor from overheating and causing safety accidents. If the vibration frequency of the reactor suddenly increases and exceeds the preset threshold, the system will also issue an early warning message to remind the operator to check whether the reactor's stirring motor is faulty. The dynamic early warning threshold can be adjusted according to the real-time operating status of the equipment, thereby improving the accuracy and timeliness of the early warning, reducing the false alarm rate, avoiding unnecessary downtime and maintenance, and also helping to detect potential faults in a timely manner and prevent accidents.

[0042] Step S109, ensure data quality through data cleaning and preprocessing, use recursive least squares method to identify parameter associations and potential failure modes, set the sliding time window length according to equipment characteristics and data cycle, and obtain data that more accurately reflects the current equipment operating conditions.

[0043] By cleaning and preprocessing the raw data, removing outliers and noise data, filling missing values, and standardizing and normalizing the data, the data quality and availability are improved. After the data preprocessing is completed, for different types of equipment, according to their working characteristics and data collection cycle, set a suitable sliding time window, slice and aggregate the data within the time window, and extract the time domain and frequency domain features that reflect the current working conditions of the equipment, such as mean, variance, peak-to-peak value, spectral entropy, etc. The extracted features are analyzed and screened by using indicators such as Pearson correlation coefficient and maximum information coefficient, and the feature subset with the greatest correlation with the equipment fault state is selected. The support vector machine SVM is used to establish a classification model between equipment parameters and fault modes, and the model hyperparameters are optimized through grid search and cross-validation to improve the accuracy of fault diagnosis. During the model training process, the weight coefficient of each feature is recorded. The larger the weight, the greater the contribution of the feature to the classification decision, which helps to understand the internal mechanism of the fault. When the model is applied, the statistical indicators such as the root mean square value and kurtosis of the equipment vibration signal are calculated to determine whether the equipment working condition has changed significantly. If the indicator exceeds the preset threshold, it is considered that the operating condition has changed, and the length of the sliding time window is dynamically shortened to improve the sensitivity and real-time performance of fault diagnosis. The fault diagnosis model is incrementally updated using the online sequential extreme learning machine OS-ELM algorithm. Whenever new equipment operation data is collected, it is automatically added to the training set, the network connection weights are updated, the dynamic changes in equipment conditions are adapted, and the model performance is continuously improved. By performing trend extrapolation and similarity analysis on the historical operation data of the equipment, the degradation trend of the health status of the equipment is predicted and the remaining service life is estimated. When the predicted remaining life is lower than the preset threshold, the multi-objective optimization algorithm is used to generate the best maintenance plan based on factors such as the importance of the equipment, maintenance cost, and consequences of failure, and to formulate spare parts replacement and maintenance strategies to minimize maintenance costs while ensuring equipment reliability and improve the economy of equipment management.

[0044] Exemplarily, data cleaning and preprocessing are the basis of fault warning and diagnosis. There may be problems such as outliers caused by sensor failure, interference caused by environmental noise, and missing data in the original data. For example, the reactor temperature sensor of a chemical plant occasionally has peak readings that are far beyond the normal range, which may be caused by sensor failure or external interference. In order to improve data quality, these outliers need to be eliminated. Commonly used methods include statistical methods, such as the 3σ criterion, which regards data exceeding the mean ±3 times the standard deviation as outliers. In addition, model-based methods can also be used, such as using Kalman filtering to smooth the data and remove noise interference. For missing values, interpolation, mean filling or model-based prediction methods can be used to complete them. Data standardization and normalization can eliminate the influence of dimensions between different features, such as converting data of different dimensions such as temperature and pressure into dimensionless data, avoiding certain features from having too much influence on the model, and improving the accuracy and stability of the model. For different types of equipment, it is necessary to set a suitable sliding time window according to their working characteristics and data acquisition cycle. For example, for rotating mechanical equipment, such as motors and pumps, the vibration signals change quickly, so a shorter time window can be set, such as a few seconds or minutes. For some equipment that changes slowly, such as reactors and storage tanks, a longer time window can be set, such as a few hours or days. Within the time window, the data is sliced ​​and aggregated to extract the time domain and frequency domain features that reflect the current working conditions of the equipment. Time domain features include mean, variance, peak-to-peak value, root mean square value, kurtosis, etc., which reflect the amplitude and fluctuation of the signal. Frequency domain features include spectrum, spectrum entropy, etc., which reflect the frequency components and energy distribution of the signal. These features can effectively describe the operating status of the equipment and provide a basis for fault diagnosis. Feature selection is one of the key steps in fault diagnosis. By analyzing and screening the extracted features, redundant features and irrelevant features can be removed, the complexity of the model can be reduced, and the diagnostic efficiency and accuracy can be improved. For example, the Pearson correlation coefficient can be used to measure the linear correlation between the feature and the fault state, and the features with higher correlation can be selected. The maximum information coefficient can measure the nonlinear correlation between the feature and the fault state, and more comprehensively reflect the importance of the feature. Assuming that the analysis shows that the mean, variance and spectral entropy of the reactor temperature have a high correlation with the fault state, while other features such as pressure and stirring motor current have a low correlation with the fault state, the mean, variance and spectral entropy of the temperature can be selected as feature subsets to build a fault diagnosis model. Support vector machine (SVM) is a commonly used fault diagnosis classification model. It can effectively handle high-dimensional data and nonlinear problems. Through methods such as grid search and cross-validation, the hyperparameters of the SVM model, such as kernel function type, penalty coefficient, etc., can be optimized to improve the generalization ability and diagnostic accuracy of the model.During the model training process, the weight coefficient of each feature is recorded to analyze the contribution of different features to fault diagnosis. For example, the weight coefficient of the temperature mean is large, indicating that the temperature mean contributes more to fault diagnosis, which helps to understand the internal mechanism of the fault. When the model is applied, the equipment working condition needs to be monitored in real time. For example, by calculating the root mean square value, kurtosis and other statistical indicators of the equipment vibration signal, it is judged whether the equipment working condition has changed significantly. If the indicator exceeds the preset threshold, it is considered that the working condition has changed, and the model parameters need to be dynamically adjusted or corresponding measures need to be taken. For example, the length of the sliding time window can be shortened to improve the sensitivity and real-time performance of fault diagnosis so as to capture the abnormal state of the equipment more quickly. Online Sequential Extreme Learning Machine (OS-ELM) is an efficient online learning algorithm suitable for processing streaming data and updating models in real time. Whenever new equipment operation data is collected, the OS-ELM algorithm can automatically add the new data to the training set and update the network connection weights, so that the model can adapt to the dynamic changes of equipment working conditions and continuously improve model performance. By performing trend extrapolation and similarity analysis on the historical operation data of the equipment, the degradation trend of the health status of the equipment can be predicted and the remaining service life can be estimated. For example, time series analysis methods, such as the ARIMA model, can be used to predict the trend of the key parameters of the equipment to determine whether the equipment is in a degraded state. When the predicted remaining life is lower than the preset threshold, it is necessary to consider factors such as the importance of the equipment, maintenance costs, and consequences of failure to formulate a reasonable maintenance plan. For example, a multi-objective optimization algorithm can be used to minimize maintenance costs and improve the economy of equipment management while ensuring equipment reliability. For example, if it is predicted that the remaining life of a key equipment is about to fall below the safety threshold, a maintenance plan can be arranged in advance to avoid equipment failures that cause production interruptions and greater economic losses.

[0045] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An intelligent early warning method for equipment status based on danger perception, characterized in that: include: Acquiring device operation data, wherein the device operation data includes multi-source heterogeneous data; Extracting features from the multi-source heterogeneous data to obtain a unified feature representation; According to the unified feature representation, a correlation model between the equipment operation status and the feature parameters is constructed; Acquire the device operation data to be analyzed, and analyze the device operation data to be analyzed according to the association model to obtain the device operation status analysis result; According to the equipment operation status analysis result, combined with a preset equipment health status evaluation index system, the current health status of the equipment is evaluated to obtain an equipment health status evaluation result.

2. The device status intelligent early warning method based on danger perception according to claim 1 is characterized in that: The extracting features of the multi-source heterogeneous data to obtain a unified feature representation includes: Acquire the multi-source heterogeneous data, and process the multi-source heterogeneous data using data cleaning and preprocessing methods; For the processed multi-source heterogeneous data, the principal component analysis method is used to reduce the dimension and extract key feature parameters; The extracted key feature parameters are input into the autoencoder network, and a unified feature representation of the multi-source heterogeneous data is obtained through unsupervised learning.

3. The device status intelligent early warning method based on danger perception according to claim 1 is characterized in that: The step of constructing a correlation model between the equipment operation status and the characteristic parameters according to the unified characteristic representation includes: Acquire historical equipment operation data, and extract characteristic parameters of the historical equipment operation data according to the unified characteristic representation; Obtaining the equipment operation status label corresponding to the historical equipment operation data; The characteristic parameters are used as input and the equipment operation status labels are used as output to train a support vector machine model to obtain an association model between the equipment operation status and the characteristic parameters.

4. The device status intelligent early warning method based on danger perception according to claim 1 is characterized in that: According to the equipment operation status analysis result, combined with a preset equipment health status evaluation index system, the current health status of the equipment is evaluated to obtain the equipment health status evaluation result, including: Obtain a preset equipment health status evaluation indicator system, the indicator system including multiple evaluation dimensions and corresponding indicator items; In combination with the equipment operation status analysis result, index items related to the current equipment operation condition are extracted from the index system as evaluation indicators; For each evaluation indicator, determine the health status of the indicator based on its threshold range and actual value; The health status of each evaluation indicator is comprehensively considered and the weighted average method is adopted to obtain the current health status evaluation result of the equipment.

5. The device status intelligent early warning method based on danger perception according to claim 1, characterized in that: The device status intelligent early warning method based on danger perception also includes: Acquire real-time data during the operation of the equipment, and dynamically update the association model using an online incremental learning method to adapt to changes in equipment operating conditions; The equipment operating status is predicted based on the updated correlation model. When the prediction result exceeds the preset threshold, an early warning is triggered and an alarm message is generated.

6. The device status intelligent early warning method based on danger perception according to claim 1, characterized in that: Before constructing the association model between the equipment operation status and the characteristic parameters, the method further includes: Based on the unified feature representation, a frequent pattern mining algorithm is used to discover association rules and time series patterns in multi-source heterogeneous data; The mined association rules and time series patterns are used as prior knowledge to guide the construction of association models.

7. The device status intelligent early warning method based on danger perception according to claim 4 is characterized in that: After the health status of each evaluation index is comprehensively evaluated by a weighted average method to obtain the current health status evaluation result of the device, the following steps are also included: Obtain equipment operation data; Continuously collect real-time data from various sensors, and perform data cleaning and preprocessing to remove outliers and noise; Compare the equipment health status assessment result with the preset health level threshold to determine the current health level of the equipment; According to the health level and importance of equipment, equipment maintenance priority plan and maintenance strategy are automatically generated to guide the development of equipment operation and maintenance work.

8. The device status intelligent early warning method based on danger perception according to claim 4 is characterized in that: Obtain a health status assessment indicator system, including: Obtain multiple indicators as evidence nodes; For the indicator system, a hierarchical analysis method is used to determine the weight of each of the evidence nodes, and the weight is used as the credibility of the evidence node; Calculating the Pearson correlation coefficient between different evidence nodes, and using the correlation coefficient as the correlation strength between the evidence nodes; Constructing an evidence reasoning network according to the credibility of the evidence nodes and the association strength; In the evidence reasoning network, a Bayesian network reasoning algorithm is used to comprehensively analyze the support of each evidence node for different health states.

9. The device status intelligent early warning method based on danger perception according to claim 7, characterized in that: The obtaining of equipment operation data includes: Preprocessing the equipment operation data to remove outliers and noise and to complete missing values; According to the type and working characteristics of the equipment, a sliding time window is set, and the pre-processed equipment operation data is sliced ​​and aggregated within the sliding time window; For the aggregated data, extract its time domain features and frequency domain features, wherein the time domain features include mean, variance and peak-to-peak value, and the frequency domain features include spectrum entropy; The Pearson correlation coefficient and maximum information coefficient are used to analyze the importance of the extracted features and select the feature subset with the greatest correlation with the equipment fault status; Using a support vector machine, a classification model between the feature subset and the equipment failure mode is established; Obtaining a weight coefficient corresponding to each feature in the classification model, and determining a feature whose weight coefficient is greater than a preset threshold as a key factor causing equipment failure; Monitor the equipment operating status online and calculate the RMS value and kurtosis of the currently collected vibration signal; If the RMS value and kurtosis exceed the preset range, it is determined that the equipment operating condition has changed significantly, and the length of the sliding time window is dynamically shortened; Adding newly collected operation data of the equipment to the training set, and incrementally updating the classification model using an online sequential extreme learning machine algorithm; Predict the remaining useful life of the equipment based on the updated classification model; If the remaining service life is lower than a preset threshold, an equipment maintenance plan is generated through a multi-objective optimization algorithm based on factors such as equipment importance, maintenance cost and failure consequence.

Images